Rainfall erosivity in catchments contaminated with fallout from the Fukushima Daiichi nuclear power plant accident

www.hydrol-earth-syst-sci.net/20/2467/2016/ doi:10.5194/hess-20-2467-2016

© Author(s) 2016. CC Attribution 3.0 License.

Rainfall erosivity in catchments contaminated with fallout from the

Fukushima Daiichi nuclear power plant accident

J. Patrick Laceby1, Caroline Chartin2, Olivier Evrard1, Yuichi Onda3, Laurent Garcia-Sanchez4, and Olivier Cerdan5 1_{Laboratoire des Sciences du Climat et de l’Environnement (LSCE), Unité Mixte de Recherche 8212}

(CEA-CNRS-UVSQ/IPSL), Université Paris-Saclay, Gif-sur-Yvette, France

2_{Georges Lemaître Centre for Earth and Climate Research, Earth and Life Institute, Université catholique de Louvain,} Louvain, Belgium

3_{Graduate School of Life and Environmental Sciences, Center for Research in Isotopes and Environmental Dynamics} (CRIED), University of Tsukuba, Tsukuba, Japan

4_{Laboratoire de Biogéochimie, Biodisponibilité et Transferts de Radionucléides, IRSN/PRP-ENV/SERIS/L2BT,} Cadarache, France

5_{Bureau de Recherches Géologiques et Minières, Orléans, France} Correspondence to:J. Patrick Laceby (placeby@lsce.ipsl.fr)

Received: 10 July 2015 – Published in Hydrol. Earth Syst. Sci. Discuss.: 30 July 2015 Revised: 29 April 2016 – Accepted: 25 May 2016 – Published: 23 June 2016

Abstract.The Fukushima Daiichi nuclear power plant (FD-NPP) accident in March 2011 resulted in the fallout of significant quantities of radiocesium over the Fukushima region. After reaching the soil surface, radiocesium is quickly bound to fine soil particles. Thereafter, rainfall and snowmelt run-off events transfer particle-bound radiocesium downstream. Characterizing the precipitation regime of the fallout-impacted region is thus important for understand-ing post-deposition radiocesium dynamics. Accordunderstand-ingly, 10 min (1995–2015) and daily precipitation data (1977– 2015) from 42 meteorological stations within a 100 km ra-dius of the FDNPP were analyzed. Monthly rainfall erosiv-ity maps were developed to depict the spatial heterogene-ity of rainfall erosivheterogene-ity for catchments entirely contained within this radius. The mean average precipitation in the region surrounding the FDNPP is 1420 mm yr−1 (SD 235) with a mean rainfall erosivity of 3696 MJ mm ha−1h−1yr−1 (SD 1327). Tropical cyclones contribute 22 % of the pre-cipitation (422 mm yr−1) and 40 % of the rainfall erosiv-ity (1462 MJ mm ha−1_h−1_yr−1 _{(SD 637)). The majority of} precipitation (60 %) and rainfall erosivity (82 %) occurs be-tween June and October. At a regional scale, rainfall ero-sivity increases from the north to the south during July and August, the most erosive months. For the remainder of the year, this gradient occurs mostly from northwest to

south-east. Relief features strongly influence the spatial distribu-tion of rainfall erosivity at a smaller scale, with the coastal plains and coastal mountain range having greater rainfall ero-sivity than the inland Abukuma River valley. Understanding these patterns, particularly their spatial and temporal (both inter- and intraannual) variation, is important for contextu-alizing soil and particle-bound radiocesium transfers in the Fukushima region. Moreover, understanding the impact of tropical cyclones will be important for managing sediment and sediment-bound contaminant transfers in regions im-pacted by these events.

1 Introduction

Radiocesium is rapidly bound to fine soil particles and remains stored within the top 0–5 cm of the soil profile (Tamura, 1964; Sawhiney, 1972; Kato et al., 2012). As ra-diocesium is adsorbed to soil particles in the upper layer of the soil profile, soil erosion and particle transport processes are considered to be the main processes responsible for trans-ferring radiocesium downstream (Saito and Onda, 2015; Ya-mashiki et al., 2014; Evrard et al., 2015). Rainfall drives erosion by detaching particles from the soil surface, primar-ily through the kinetic energy of raindrops (Wischmeier and Smith, 1958; Renard and Freimund, 1994). The potential of rainfall to erode soils is referred to as rainfall erosivity. Quan-tifying rainfall erosivity is important to determine the cli-matic vulnerability of a region, like Fukushima, to rainfall-driven soil erosion (Mello et al., 2013).

The Universal Soil Loss Equation (USLE) (Wischmeier and Smith, 1978) and its revised edition (RUSLE) (Renard et al., 1997) are two of the most broadly adopted soil ero-sion modelling frameworks (Oliveira et al., 2013). The USLE predicts average annual soil loss with six factors pertaining to soil, topography, vegetation, management and rainfall ero-sivity, known as theRfactor. TheRfactor estimates rainfall erosivity as a combined function of event frequency, inten-sity and rainfall amount (Wischmeier and Smith, 1958). It sums event-based erosivity individually calculated through multiplying each storm’s total kinetic energy with its maxi-mum measured 30 min rainfall intensity. To incorporate vari-ations in interannual rainfall, the R factor is averaged over long temporal periods (∼20 years) (Renard and Freimund, 1994; Wischmeier and Smith, 1978).

When all other factors in the (R)USLE are held constant, there is a proportional relationship between soil loss and the R factor (Lu and Yu, 2002; Wischmeier and Smith, 1958, 1978). The utility of the R factor is reflected in its global application (McFarlane et al., 1986; Ma et al., 2014; Oliveira et al., 2013; Moore, 1979; Panagos et al., 2015), including Asia in general (Lee and Heo, 2011; Shamshad et al., 2008; Yu et al., 2001) and Japan in particular (Shiono et al., 2013). There are limitations to the application of theRfactor and the USLE (Kinnell, 2010). Although the USLE andRfactor were developed from plot-scale soil erosion research (Wis-chmeier and Smith, 1978), they have been applied at the catchment scale (Wilkinson et al., 2009; Kitamura et al., 2014; Belyaev et al., 2005). A debate about the applicability of the USLE within catchment-scale modelling frameworks and the universality of the R factor is beyond the scope of this current research. What is important is that theR factor and a thorough characterization of the precipitation regime are useful to contextualize local soil erosion, which repre-sents the first step of potential downstream particle-bound contaminant transfers.

Research in the Fukushima region has clearly indicated that precipitation drives downstream radiocesium migration (Evrard et al., 2015). In fact, the majority of soil erosion and concomitant radiocesium migration was reported to coincide

with extreme rainfall events (Mouri et al., 2014; Nagao et al., 2013; Onishi et al., 2014). For example, Yamaguchi et al. (2014) reported that most sediment-bound radiocesium migration occurs in flood events that may only occur 1 to 2 times per year. Tropical cyclones in this region likely trans-fer significant quantities of radiocesium from hillslopes to the Pacific Ocean (Evrard et al., 2015).

As improved estimates of rainfall erosivity result in more accurate modelling results (Renard et al., 1991; Lee and Heo, 2011), a comprehensive examination of rainfall erosivity, in-cluding the impact of tropical cyclones on erosivity, may im-prove our understanding of sediment and radiocesium dy-namics in the Fukushima region. Therefore, the objective of this research is to evaluate the spatial and temporal (monthly and interannual) precipitation and rainfall erosivity in gen-eral, and from tropical cyclones in particular, for catchments contained entirely within a 100 km radius of the FDNPP ac-cident. Rainfall erosivity maps (monthly and annual) are pro-vided for the Fukushima region, along with precipitation and rainfall erosivity data for all events selected by the RUSLE criteria and information summarizing tropical cyclone events impacting the area since 2011, in the Supplement.

2 Methods 2.1 Study area

Most of the terrestrial fallout from the FDNPP accident oc-curred over the coastal catchments of the Fukushima Prefec-ture (Fig. 1). Soil contamination > 25 kBq kg−1of134+137Cs occurred within small catchments (< 700 km2) draining to the Pacific Ocean, with the highest contamination levels in the coastal mountain range northwest of the FDNPP. The coastal mountain range is separated from the larger west-ernmost mountain range by the Abukuma River catchment (∼5200 km2) extending on a SSW–NNE direction and char-acterized by a134+137C soil contamination < 25 kBq kg−1.

Figure 1. Standard, omitted and long-term stations in the Fukushima region, along with catchments (grey lines) contained entirely within a 100 km buffer around the FDNPP (dashed line); 134+137_{Cs activities are from Chartin et al. (2013).}

2.2 Meteorological stations

All precipitation monitoring stations within a 100 km ra-dius of the FDNPP were analyzed (Fig. 1). Fourteen stations within this 100 km radius were omitted due to an operation period of less than 7 years (n=5) or operation pause during winter months (n=9). Up to 21 years were available for the 10 min (1995–2015) and 39 years for daily precipitation data (1977–2015) for 42 stations. Daily precipitation was also an-alyzed from three additional long-term stations (including one located within a 115 km radius of the FDNPP) as they had similar data available from 1890 to 2015. The long-term stations were excluded from theR factor and daily rainfall calculations due to high levels of missing data (mean 33 % daily) and short temporal periods (7 years–10 min). All data were downloaded from the Japan Meteorological Agency’s (JMA) website (http://www.jma.go.jp/jma/index.html).

Daily precipitation data were examined for breakpoints and homogeneity with the RHTest program from the Mete-orological Service of Canada (Wang et al., 2010). No

sig-nificant breakpoints required correction. Duan et al. (2015) similarly did not report significant breakpoints for the JMA daily precipitation data. No homogenization analyses were performed on the 10 min precipitation data due to the dif-ficulties in assessing heterogeneity in 10 min precipitation data and the potential to overcorrect the data. Shiono et al. (2013) similarly did not correct 10 min JMA data. For the daily data, there was an average of only 0.23 % of miss-ing data (SD 0.20 %, maximum 0.94 %) compared to 0.30 % (SD 0.20 %, maximum 0.78 %) for the 10 min data.

2.3 Rfactor calculations

The Rainfall Intensity Summarisation Tool (RIST) software (USDA, 2013) was used to calculate theR factor, which is a product of the kinetic energy of an event (E) and its maxi-mum 30 min intensity (I30) (Renard and Freimund, 1994):

R=1

n

n

X

j=1 mj

X

k=1

(EI30)k, (1)

where R equals the annual average rainfall erosivity in MJ mm ha−1h−1yr−1,nrepresents the number of years of data utilized, mj is the number of erosive events in year j

andEI30was calculated as

EI30= o

X

r=1 ervr

!

I30, (2)

whereerrepresents rainfall energy per unit depth of rainfall in MJ ha−1mm−1,vris the volume of rainfall (mm) during a given time interval (r) andI30is the maximum rainfall inten-sity over a 30 min period of the rainfall event (mm h−1). For each time interval,erwas calculated as

er=0.29 h

1−0.72(−0.05ir)i_{, (3)} where ir is rainfall intensity (mm h−1) (Renard and Freimund, 1994).

To calculate the annual and monthly R factors, erosive rainfall events were summed for each monitoring station for the considered period (months or year). Two criteria within the RIST software were used to define an erosive event: (1) cumulative rainfall of a given event is > 12.7 mm; (2) ac-cumulations < 1.27 mm over a 6 h period separates events into two different periods.

To examine the impact of tropical cyclones on precipita-tion and rainfall erosivity, the best track data of all tropi-cal cyclones from the JMA Tokyo-Typhoon Center (JMA, 2016) were cross-referenced with regional events that oc-curred from 1995 to 2015. Each cyclone path and summary information was also reviewed in the Wikipedia Pacific ty-phoon database (Wikipedia, 2016). Furthermore, news re-ports on tropical cyclones in the region (from Google news searches) were also examined to provide complimentary in-formation. A database was then created that allowed for the quantification of the precipitation and rainfall erosivity that occurred during tropical-cyclone-related events. In the Northwest Pacific Ocean, tropical cyclones are referred to as tropical depressions, tropical storms and typhoons depend-ing on the sustained wind speed. Tropical cyclones are non-frontal synoptic-scale systems that have organized convec-tion along with cyclonic surface wind circulaconvec-tion. Typhoons, tropical storms and tropical depressions are all referred to as tropical cyclones in the remainder of this paper.

2.4 Rfactor spatial interpolation

Annual and monthlyRfactor data were used to derive maps of annual and monthly rainfall erosivity at a resolution of 250 m over the eastern part of the Fukushima Prefecture. Rather than interpolation by kriging, a regression approach, based on relationships established between theRfactor and spatially distributed covariates, was used to produce these maps in order to identify the environmental covariates that are driving the spatial distribution of rainfall erosivity.

The covariates used for the spatial interpolation of annual and monthlyR factors over the study area are described be-low.

– Climatic data (i.e. monthly mean precipitation and mean temperature) from the WorldClim database reported for the period 1950–2000 and with a 30 arcsec resolution (equivalent to 1 km resolution) (Hijmans et al., 2005) were used for the monthly models (Panagos et al., 2015). From the monthly data, annual mean precipita-tion and mean temperature were derived to be included as covariates in the annual R factor model. These co-variates are expected to impact the spatial distribution of rainfall erosivity.

– Based on their known influence on spatial patterns and intensities of rainfall, different morphometric attributes have been included as covariates in the mapping pro-cedure, as proposed by Daly et al. (2002). Precipita-tion varies strongly with elevaPrecipita-tion (Barry and Chorley, 2009). Conceptually, relief features can induce different precipitation–elevation gradients depending partially on their ability to block and uplift moisture-bearing air (Daly, 2002). In mountainous areas, the steepest fea-tures oriented perpendicular to the air flow are more likely to produce a greater elevation–precipitation

gra-dient than gentle sloping areas parallel to the air flux, such as coastal plains. Moreover, hillslopes experience different climate regimes depending on their orienta-tion, relative to air currents at large scales (i.e. leeward and windward sides of mountains), and solar radiation. Areas located at similar elevations may thus have differ-ent precipitation intensities.

Accordingly, a 90 m resolution digital elevation model (DEM) was obtained from Shuttle Radar Topography Mission (SRTM) of NASA (Jarvis et al., 2008). Con-sidering the local relief in the fallout-affected region (Fig. 1), the DEM was smoothed with a Gaussian fil-ter to two wavelengths of 5 and 20 km. From the 20 km wavelength smoothed DEM, slope and aspect were de-rived. The slope layer was used in the modelling pro-cedure to account for potential impact of large relief features on moisture-bearing air. The aspect layer de-rived from the same 20 km wavelength smoothed DEM was used to delineate the opposite sides of mountains and represent leeward and windward sides. Then, this aspect layer was inserted in the model as a categori-cal covariate (i.e. a two-value layer, one value for each leeward and windward “facets”). From the 5 km wave-length smoothed DEM, the easting and northing were computed to represent the degree to which aspect is close to the east and the north, respectively (Zar, 2010). These two layers were used as covariates in the mod-elling procedure to account for slope orientation at a smaller scale than the entire side of a mountains range at the coarser scale.

– Distance to the Pacific Ocean coast (DC is the distance to the coast) was calculated for each pixel by using the Euclidean distance tool from ArcGis 10 (ESRI, 2011) as the proximity of the ocean, a major source of moisture, may influence the spatial distribution of precipitation. Before the fitting of the models, the continuous covariates presented above were rescaled at a 250 m resolution in Ar-cGIS10 (ESRI, 2011) with bilinear interpolation, which de-termined the output value of a cell based on a weighted dis-tance average of the four nearest input cell centres. Then, the spatial variation of rainfall erosivity in the study region was modelled with generalized additive models (GAMs; Hastie and Tibshirani, 1986), implemented in the mgcv package in R (Wood, 2001). The GAM regression technique is a gen-eralization of linear regression models (GLMs) in which the coefficients may be a set of smoothing functions. GAMs con-sider the nonlinearity that may exist between the target vari-able (Y) and covariates (X), providing more flexibility to the model fitting than a GLM.

function of the selected predictors:

g[µ(Y )] =α+f1X1+f2X2+. . .+fpXp, (4) whereYis the response variable,X1,X2, . . .Xprepresent the

covariates and the fi is the smooth (non-parametric)

func-tions.

Based on monthly and annual mean values for each of the 42 stations, 13 GAMs were fitted to spatially model the R factor over the study area (i.e. one model per month and one model for the year). A Gaussian distribution model in-corporated the conditional mean µ(Y) and, due to the pre-dominant logarithmic distribution of the monthly and annual data, a log-linear link function g(µ)=log(µ) was imple-mented. The smoothing functions of the models were built using regression splines fitted by penalized maximum like-lihood to avoid over-fitting (Wood, 2001). An extra penalty was added to each smoothing term so that each could poten-tially be set to zero during the fitting process, especially in the case of multi-collinearity or multi-concurvity. The inter-action of geographical coordinates was then added to each model (as a two-dimensional spline on latitude and longi-tude) to incorporate spatial trends of the target variable at the regional scale.

A model including all the covariates was first developed (the “full” model), based on each of the total data sets (13 data sets ofn=42). Next, a backward stepwise procedure was applied to select the appropriate covariates, where each of the covariates was sequentially removed from the full model (Poggio et al., 2013; de Brogniez et al., 2015). The dif-ference (1AIC) between AIC values (Akaike’s information criterion; Akaike, 1974) was calculated for the “full model” and compared with AIC values obtained when removing se-quentially each of the covariates from the model in order to evaluate the influence of each covariate on the overall capa-bilities of model prediction.

Explained variance was calculated during the procedure to assess the evolution of “fitting performance” and to sup-port the decision to keep or remove a covariate into the fi-nal model. Once the fifi-nal model was obtained, a Bayesian approach was used to compute standard errors for the pre-dictions as proposed in the mgcv package. A leave-one-out cross-validation (LOOCV) procedure was applied to each fit-ted model. Predicfit-ted and observed values were compared with the calculation of the root mean square errors (RMSEs) and the coefficient of determination (R2). Finally, maps were delineated according to the limits of the target area, i.e. en-tire hydrological basins contained within the 100 km radius of around the FDNPP.

3 Results 3.1 Precipitation

Mean annual precipitation ranged from 1098 mm yr−1 at station 281 to 2201 mm yr−1 at station 1116 (Table 1; Fig. 2a). Mean annual precipitation was 1420 mm yr−1 (SD 235 mm yr−1) (Fig. 2a) with a coefficient of variation (CV) of 17 %. Between 1977 and 2015, regional mean an-nual precipitation ranged from 869 mm in 1984 to 1844 mm in 2006 (Fig. 3a) with a CV of 15 %. This interannual vari-ation in precipitvari-ation was maintained over a longer period (Fig. 4), with a mean annual precipitation for the long-term stations (126 yr) of 1159 mm yr−1(SD 187 mm) with a 16 % CV. The 5-year running average for these long-term stations demonstrates that the region experiences prolonged wet and dry periods that last approximately 5–10 years. Note that these long-term stations plot near or below 1 standard devia-tion on the annual precipitadevia-tion mean for all of the stadevia-tions, and therefore they may not be indicative of all precipitation trends in the region.

The majority of precipitation (60 %) occurs between June and October (Fig. 5a) coinciding with a wet season that in-cludes tropical cyclones. In fact, 39 % of the precipitation oc-curred between July and September alone, compared to only 22 % between March and May and 17 % between November and February. The monthly precipitation also varied spatially. For example, the mean monthly maximum precipitation was 318 mm m−1at site 1116 in July compared to 256 mm m−1 for site 281 in July.

The highest daily maximum precipitation recorded was 607 mm in 1998 at site 326, compared to the lowest daily maximum precipitation of 38 mm in 1984 at site 294 (Fig. 6a). The mean daily maximum precipitation for all the stations and every year was 115 mm (SD 27 mm). The high-est mean daily maximum precipitation for all the stations (169 mm) was recorded in 2011 compared to a minimum of 62 mm in 1984. Although the maximum daily precipitation is not an optimal indicator of precipitation depth due to the probability of events occurring over multiple days, the year of the FDNPP accident (2011) had the highest daily maxi-mum precipitation as a result of Typhoon Roke.

3.2 Rainfall erosivity

The annualRfactor ranged from 1972 MJ mm ha−1h−1yr−1 at station 290 to 8274 MJ mm ha−1_h−1_yr−1_{at station 326} (Table 1 – Fig. 2b). The meanRfactor for the 42 stations was 3696 MJ mm ha−1h−1yr−1 (SD 1327) with a CV of 36 %. The CV was double the corresponding CV for annual aver-age precipitation indicative of more spatial variability.

Table 1.Station coordinates along with the mean and standard deviation (SD) of annual precipitation,Rfactor and kinetic energy (KE).

Station Latitude Longitude Annual precipitation Rfactor KE (mm) (MJ mm ha−1h−1yr−1) (MJ ha−1yr−1) Mean SD Mean SD Mean SD

254A,4 38.178 140.635 1429 302 3295 3045 167 61 256 38.015 140.612 1292 243 2497 1296 149 42 257 38.025 140.858 1256 259 2922 1313 148 38 279 37.912 140.143 1343 221 2049 572 142 28 281 37.852 140.588 1098 172 2045 792 121 26 285 37.783 140.925 1350 266 3859 1928 171 48 288∗ 37.638 140.983 1333 272 3819 1527 173 39 290 37.552 140.108 1265 242 1972 1029 124 35 291 37.583 140.430 1176 219 2337 1206 136 33 294 37.435 140.577 1168 210 2718 1355 141 36 295 37.492 140.965 1493 299 4327 1694 196 43 299 37.368 140.330 1154 176 2465 982 130 30 300∗,D 37.347 141.017 1532 295 4249 1517 203 40 304 37.287 140.625 1231 200 2984 1361 145 31 307 37.147 140.460 1280 205 3405 1530 157 34 310∗ 37.065 140.877 1426 281 4425 2105 193 49 312 36.938 140.408 1399 234 4342 2145 182 46 314∗ 36.868 140.637 2001 380 7177 2405 274 55 315 36.833 140.772 1472 247 4126 1629 185 42 31610 36.778 140.345 1419 204 5075 2370 193 39 326 37.123 140.035 1954 373 8274 8491 279 78 1011 36.580 140.645 1474 232 4406 2002 191 37 1034 37.233 141.000 1568 280 4530 1545 205 45 11163 37.668 140.260 2201 355 5056 2385 262 57 1127C,6 37.008 140.737 1435 285 3636 1263 178 46 1129 37.337 140.808 1444 250 4193 2115 185 42 1130 37.695 140.747 1307 245 2989 1563 157 41 113210 38.003 140.207 1254 199 2350 1380 127 34 1150∗,10 37.560 140.753 1341 239 3470 1369 165 35 1173∗,10 36.778 140.482 1437 234 4432 1981 191 42 122010 37.932 140.778 1262 245 2878 1234 149 40 1268∗,8 37.288 140.218 1366 256 3680 2595 170 50 1274∗,8 37.207 140.750 1440 272 4156 1812 181 43 12829 37.722 140.058 1762 316 3351 1764 200 49 12939 37.892 140.437 1363 178 2361 916 144 29 12949 37.277 140.063 1568 281 3914 2567 190 51 1298∗,8 37.827 140.728 1462 270 3765 2209 175 46 1386∗,7 37.388 140.090 1311 231 2726 1569 142 39 1421∗,5 36.743 140.593 1851 284 6448 2422 249 50 1464B,2 38.138 140.917 1129 174 2742 1522 133 27 1466B,2 37.227 140.427 1232 213 3150 1055 146 33 1564A,1 38.127 140.680 1356 170 2647 1030 154 26

∗_{No temperature data available.}

ForRfactor,A_{10 years of data available,}B_{13 years,}C_{14 years and}D_{18 years.}

For precipitation analyses,110 years available,213 years,321 years,428 years,530 years,632 years,733 years,836 years,937 years and10_{38 years.}

with a CV of 34 %. The CV again was more than double that of the annual precipitation.

The highest monthly R factors occurred in the summer months and early fall (Fig. 5b), coinciding with the wet

Figure 2.Total (grey) and typhoon-related (red) mean annual pre-cipitation(a)andRfactor(b)with the solid line representing the mean and the dashed lines being 1 standard deviation from the mean. The error bars indicate 1 standard deviation from the mean for individual stations.

In particular, 64 % of the rainfall erosivity occurs between July and September. Winter (November–February) only con-tributes 6 % of the total annualRfactor with spring (March– May) contributing the remaining 12 %.

3.3 Spatial distribution of rainfall erosivity

The variance explained by the final erosivity models pro-duced by the calibration procedure (from the total data set of n=42) varied from 82.5 % for September to 97.7 % for April and was 95.5 % for the annual R factor model (Ta-ble 2). Predicted values produced with the LOOCV pro-cedure for the 42 stations showed similar distributions as the observed values (Table 2) for the year and for the months between October and June. For the months between July and September, which includes the tropical cyclone season, the standard deviations of predicted values were smaller than for the observed values. For mean observed

values ranging from 778.7 to 799.7 MJ mm ha−1h−1m−1 between July and September, the RMSE varied between 165.3 and 313.6 MJ mm ha−1_h−1_m−1 _{with a R}2 _between 0.56 and 0.67. The model fitted for August demonstrated a bias by underestimating theR factor (mean observed val-ues=778.7 MJ mm ha−1h−1m−1 and mean predicted val-ues=755.2 MJ mm ha−1h−1m−1).

For the months between October and June, and for the year, the differences between mean observed and predicted values remained relatively low relative to corresponding ob-served values and their variability. For these periods, theR2 produced by the validation procedure varied between 0.62 in April and 0.86 in October, with aR2of 0.77 for the an-nual model. In general, the anan-nual map and the months with higher erosivity had lower R2 values. Contrarily, months with lower erosivity often had higherR2. This is possibly due to localized impacts of high precipitation events.

For all of the monthly models, the categorical variable facets, was selected as one of the most influential explanatory covariates, whereas facets were less influential in the annual model (Table 3). The selection of facets highlights how oppo-site sides of the mountain ranges in this region have different rainfall erosivity regimes.

For 10 of the 13 models, mean precipitation, for the con-sidered period, was selected as a significant explanatory co-variate. Mean temperature was selected by the procedure for the three models for which mean precipitation was not re-tained as significant covariates (i.e. February, November and December). Mean temperature and mean precipitation was selected for the annual R factor model. According to the 1AIC, mean precipitation and mean temperature capture the mostRfactor spatial variability within the final models (Ta-ble 3). Easting and/or northing was selected in all the other models, except July and September, accounting for the influ-ence of hillslope aspects at smaller scales than entire moun-tain sides. The selection of elevation, slope and/or the dis-tance to the coast as significant explanatory variables varied in some of the final models.

Although elevation, slope and distance to the coast may influence precipitation patterns and intensities, their incon-sistent selection in the final models does not mean that they potentially do not influence rainfall erosivity. Covariates have the potential not to be retained in the final model when there may be multi-collinearity or multi-concurvity between co-variates. The non-selection of covariates signifies that other covariates likely have a similar, though stronger, mathemat-ical relationships with the target variable. To achieve parsi-mony, only selected variables from the original set of covari-ates are included in the final model.

south-southeast-Figure 3.Annual (grey) and typhoon-related (red) mean annual precipitation(a)and corresponding standardized anomaly index (SAI –b) and the annual (grey) and typhoon-related (red)Rfactor(c)and theRfactor’s SAI(d). The grey solid line represents the regional mean (including typhoons) and the grey dashed lines represent 1 standard deviation from the total mean, whereas the red dashed line represents the mean contribution from typhoon events. The blue (total) and red (typhoon) solid lines depict the 5-year running average.

Figure 4.Mean annual precipitation(a)and the standardized anomaly index (SAI –b) for three long-term rainfall stations including mean annual precipitation (solid line), the standard deviation (dashed line) for the entire period and the 5-year running average (blue line).

oriented side of the westernmost mountain range bordering the Abukuma valley has lower rainfall erosivity than the easternmost side of this valley. The CV of modelling errors is rather low, predominantly under 10 % in the study area (Fig. 7b).

In July and August, the most erosive months of the year, rainfall erosivity increases mainly from the north to the south of the region, while during the rest of the year this gradient occurs mostly from northwest to southeast. The spatial pat-terns relative to the relief features are also observed in the monthly rainfall erosivity distributions with the coastal plains and coastal mountain range receiving higher rainfall erosivity

than the parallel Abukuma River valley. The maps of errors associated with each monthlyRfactor models are available in the Supplement (Fig. S1). The coastal plain and coastal mountainous ranges, including the FDNPP and a large part of the contamination plume with high levels of radiocesium contamination, have greater rainfall erosivity than the less contaminated Abukuma River valley.

3.4 Tropical cyclones

Table 2. Observed and predicted mean and standard deviation (SD) for monthly (in MJ mm ha−1h−1m−1) and annual R factors (in MJ mm ha−1h−1yr−1) with GAM performance parameters (RMSE: root mean square error;R2: coefficient of determination).

Month or year ObservedR PredictedR Variance Model performance explained (%) (LOOCV) Mean SD Mean SD RMSE R2

Year 3696 1327 3696 1293 96 640 0.77 January 43 26 44 25 90 16 0.77 February 24 16 23 15 90 9 0.68 March 64 31 64 30 93 13 0.83 April 160 104 165 97 92 66 0.62 May 206 160 214 163 98 75 0.80 June 312 124 312 116 88 66 0.72 July 800 283 800 244 83 167 0.65 August 779 469 755 305 92 314 0.56 September 785 287 781 263 83 165 0.67 October 368 247 374 237 95 92 0.86 November 73 28 72 25 90 12 0.80 December 83 66 84 62 93 33 0.76

Table 3.Results from the backward stepwise procedure used to select covariates to include in final GAMs for annual and monthlyRfactor maps computation. The values (1AIC) correspond to the difference in AIC scores obtained when the particular covariate is dropped from the “full” model, i.e. the model containing all the covariates. Scores in bold indicate which covariates were selected in the final models.

Model 1AIC

Elevation Slope DC∗ Easting Northing Rainfall Temperature Facets

Annual 8.7 1.0 10.8 12.1 2.8 13.6 10.1 5.6

January 0.0 6.1 2.1 0.7 0.0 8.3 0.0 4.0

February 2.2 0.0 0.1 2.7 2.6 0.2 4.6 8.6

March 0.0 9.9 0.0 1.3 0.1 5.3 0.0 3.2

April 3.7 0.0 17.6 0.0 0.5 17.3 0.0 14.0

May 0.1 8.5 0.0 0.7 4.5 8.4 0.0 10.9

June 0.4 0.3 0.0 0.2 1.4 8.6 0.0 12.6

July 0.0 0.3 5.5 0.0 0.0 13.2 0.0 13.8

August 0.7 1.6 0.0 0.0 8.6 8.8 0.1 7.8

September 0.0 0.4 5.5 0.0 0.0 13.2 0.0 13.8

October 0.4 4.8 1.2 0.0 5.7 16.3 0.0 19.8

November 0.0 26.3 2.7 3.0 0.2 0.2 11.2 8.8

December 0.3 0.2 0.0 3.2 0.4 0.6 11.7 5.5

∗_{Distance to coast.}

This contribution varies from only 5 % (68 mm yr−1) in 2003 to 35 % (639 mm yr−1) in 1998. In September, 60 % of the precipitation is derived from tropical-cyclone-related events (119 mm m−1) compared to 43 % in October (64 mm m−1), 28 % in August (49 mm m−1) and 25 % in July (47 mm m−1). During the remainder of the year there is less than a 15 mm m−1 contribution from tropical cyclones (< 11 % of the month precipitation totals) (Fig. 5a).

Although tropical cyclones only contribute 22 % of the annual precipitation in the region, they contribute 40 % of the annual rainfall erosivity (1462 MJ mm ha−1h−1yr−1). This contribution varies from 6 % (142 MJ mm ha−1_h−1_yr−1_{) in 1995 to 64 % in}

Figure 5.Total (grey) and typhoon-related (red) mean monthly rain-fall(a)andRfactor(b)with error bars indicating 1 standard devi-ation from the mean, the solid line representing the monthly mean and dashed lines representing 1 standard deviation from the mean. The mean monthly temperature is plotted (red line) along with mean monthly rainfall(a).

4 Discussion 4.1 Precipitation

The location and intensity of annual tropical cyclones influ-ence the spatial and temporal (intra- and interannual) varia-tion in precipitavaria-tion in the Fukushima region. The majority of precipitation occurs in the summer months, the result of the high humidity and seasonal tropical cyclone activity. Long-term climatic oscillations (e.g. El Niño) also influence the precipitation patterns, illustrated by the 5- to 10-year running average reported in the long-term analyses (Fig. 4). Duan et al. (2015) reported periodic variations in extreme precipita-tion though at frequencies of 2–3 and 5–13 years.

The average precipitation in the Fukushima region (1420 mm yr−1) is∼250 mm less than the mean annual pre-cipitation in Japan (1684 mm yr−1) (Kim et al., 2010). Along the western and southeastern coast of Japan, annual precipi-tation may be greater than 3000 mm yr−1(Kim et al., 2010).

In the southeast of Japan, the elevated annual precipitation is related to an increasingly tropical climate progressing south-wards and the increased occurrence of tropical cyclones in these warmer climates with higher precipitation intensities during summer months.

4.2 Rainfall erosivity

Rainfall erosivity varied spatially and temporally (monthly and interannually) in the Fukushima region. Rainfall erosiv-ity has similarly varied in other regions of the world, e.g. the Mediterranean region (Capolongo et al., 2008) and the Middle East (Eltaif et al., 2010). Although researchers have demonstrated relationships between precipitation and rainfall erosivity (Ma et al., 2014), in the Fukushima region the rela-tionship was positive, though not significant (r2=0.39). The lack of significant relationship was likely the result of rain-fall erosivity varying more than precipitation. There are also possibly decadal variations in rainfall erosivity, similarly to precipitation, that require long-term analyses to investigate.

There have been a range ofR factors reported for Japan and the Fukushima region. In the Fukushima region, between 17 July 2011 and 18 November 2012, Yoshimura et al. (2015) calculated anRfactor of 3677 MJ mm ha−1h−1(SD 463) for three sites near Kawamata town,∼13 km from station 1130. Although station 1130 recorded a similar R factor for this time period (3853 MJ mm ha−1h−1), theR factor recorded at this station was 40 % lower than the regional average for this period (6559 MJ mm ha−1h−1, SD 3232). This demon-strates the potential impact of the spatial heterogeneity on rainfall erosivity and limitations of extrapolating data from one sampling point to the entire region.

Yamaguchi et al. (2014) quantified an average annual R factor of 3366 MJ mm ha−1h−1yr−1 for 23 JMA sta-tions in this region for a 10-year period between 2001 and 2011. For a sensitivity analysis, these authors also mod-elled a maximum case of 4217 MJ mm ha−1h−1yr−1 and a minimum case of 1986 MJ mm ha−1h−1yr−1. The aver-age annual R factor reported by Yamaguchi et al. (2014) was 9 % lower than that reported for our current research, likely the result of the longer temporal period and the high levels of rainfall erosivity in the late 1990s not be-ing included in the calculations of Yamaguchi et al. (2014). Further, there was only a 120 MJ mm ha−1h−1yr−1 dif-ference between our minimum cases, though our maxi-mum case was 41 % higher (7159 MJ mm ha−1h−1yr−1) in 1998 than Yamaguchi et al. (2014) and again 33 % higher (6288 MJ mm ha−1_h−1_yr−1_{) than their maximum case in} 1999.

Figure 6.Mean maximum daily precipitation(a)and corresponding SAI(b)with error bars indicating 1 standard deviation from the mean for each station, along with the mean (solid) and 1 standard deviation from the mean (dashed line) for the daily rainfall analysis period; the blue solid line depicts the 5-year running average.

Figure 7.AnnualRfactor(a)and model errors depicted as the coefficient of variation (i.e. CV=(standard error/mean)·100)(b)for catch-ments entirely contained within a 100 km radius of the FDNPP.134+137Cs activities are from Chartin et al. (2013).

the Hirose River catchment (Miyagi Prefecture in north-ern Japan) (Takeuchi and Ishidaira, 2001). Shiono et al. (2013) found that rainfall erosivity varied from 721 to 35 299 MJ mm ha−1h−1yr−1 across Japan with a mean of 5130 MJ mm ha−1h−1yr−1. In our current analyses, rainfall erosivity in the Fukushima region was found to be similar to the results of Shiono et al. (2013), who spatially depicted rainfall erosivity in the Fukushima region between 2500 and 5000 MJ mm ha−1_h−1_yr−1_.

Shiono et al. (2013) also demonstrated that rainfall ero-sivity in the Fukushima region is relatively low compared to the rest of Japan. Similarly to annual precipitation, there is a rainfall erosivity gradient in Japan from north to south, particularly along the Pacific coast with rainfall erosivity in-creasing southwards from the colder temperate/subarctic re-gions in the north towards the tropical rere-gions in the south. Importantly, the rainfall erosivity of the Fukushima region is

projected, by Shiono et al. (2013), to increase significantly as a result of climate change.

The difference between Shiono et al. (2013) and our cur-rent research is the comprehensive analysis of precipitation, and rainfall erosivity at a scale relevant to the fallout from the FDNPP accident. In particular, our current research high-lights that the coastal mountain range, which received a sig-nificant amount of radiocesium fallout, has higher rainfall erosivity relative to the remainder of the fallout-impacted re-gion, with a large proportion of this rainfall erosivity being derived for tropical cyclones.

4.3 Tropical cyclones

Figure 8.MonthlyRfactor distribution for catchments contained entirely within a 100 km radius of the FDNPP.134+137Cs activities are from Chartin et al. (2013).

to range between 9000 and 14 000 MJ mm ha−1h−1yr−1 (Shamshad et al., 2008) in one study and between 1360 and 21 600 MJ mm ha−1h−1yr−1in another (Yu et al., 2001). In these regions, the majority of the annual precipitation and rainfall erosivity may be derived from only a few storms (García-Oliva et al., 1995). In the Fukushima region, this

Figure 9.Total (grey) and typhoon-related (red) monthly rainfall erosivity from 2011 to 2015 with the mean line representing mean monthly rainfall erosivity and the dashed line being 1 standard deviation from the mean for the period of 1995–2015 for the 42 rainfall stations analyzed.

4.4 Implications for radiocesium dynamics

Precipitation characteristics, such as rainfall erosivity, are fundamental inputs into a variety of approaches to modelling downstream contaminant transfers (Karydas et al., 2015; Ya-maguchi et al., 2014; Rosa et al., 1996). Although precipi-tation and rainfall erosivity alone are insufficient to directly model downstream radiocesium migration, their characteri-zation is important to improve our understanding of radioce-sium dynamics. In the Fukushima region, rainfall drives three major contaminant transfer processes: (i) wet fallout of par-ticle from the atmosphere to the surface controls the spatial distribution of deposited contaminants; (ii) rainfall splash ef-fects trigger the detachment of particle-bound contaminants; (iii) detached particles are potentially transported by over-land flow generated by rainfall when infiltration capacity is exceeded or when the soil is saturated with water. In addition,

the timing and intensity of events in general, and tropical cy-clones in particular, are likely of particular importance for understanding these and other post-fallout radiocesium dy-namics in the Fukushima region (Fig. 9).

Although the Chernobyl and Fukushima regions have similar humid continental climates (Dfb), the tropical cy-clones and other significant summer rainfall events result in a higher precipitation (∼2 times) and rainfall erosivity in the Fukushima region (∼3–4 times). In the Kharkiv re-gion of Ukraine, the mean annual precipitation is 460 mm yr1 (Nazarok et al., 2014); in Odessa it is < 500 mm yr−1 (Svetl-itchnyi, 2009) compared to 1420 mm yr−1for the Fukushima region. In Ukraine, the region within a 100 km radius from Chernobyl has a rainfall erosivity that ranges from 830 to 1250 MJ mm ha−1h−1yr−1(Larionov, 1993). Within 100 km from the FDNPP accident, theR factor is∼3 to 4 times higher (3696 MJ mm ha−1h−1yr−1). As there is pos-itive relationship between rainfall erosivity and soil loss, when other factors are held equal, higher erosion rates and therefore a more rapid radiocesium export are anticipated in the Fukushima region compared to Chernobyl, based on these differences in their climatic contexts.

R factor maps focusing on the region impacted by fallout from the FDNPP accident that depict the annual and monthly variation in theRfactor are important for research and mod-elling of radiocesium dynamics in this region. TheseRfactor maps are thus provided (data set 1). Further, all the event data (total precipitation, rainfall erosivity and kinetic energy) are included in the Supplement (data set 2). These two data sets are included in the Supplement. More research is neverthe-less required to examine the impact of different raindrop size distributions and resultant variations in kinetic energy in this region.

5 Conclusions

Rainfall erosivity is an important indicator of a region’s cli-matic vulnerability to rainfall-driven soil erosion and also the potential downstream transfer of sediment and sediment-bound contaminants. Rainfall erosivity in the Fukushima re-gion is 3696 MJ mm ha−1h−1yr−1 and the region receives 1420 mm yr−1of precipitation. Although these are below av-erage values for Japan, the rainfall erosivity is more than 3–4 times higher than in Chernobyl. The majority of precipita-tion (60 %) and rainfall erosivity (86 %) occurs between June and October. Major tropical cyclone events are responsible for 22 % of the precipitation (422 mm yr−1) and 40 % of the rainfall erosivity (1462 MJ mm ha−1h−1yr−1, SD 637). The highest areas of contamination in the coastal mountain range correspond to the contaminated area that receives high rain-fall erosivity.

TheR factor maps demonstrate that areas most contami-nated with radiocesium with the highest rainfall erosivity are located along the coastal mountain range of the Pacific Ocean west of the FDNPP. Catchments in these areas will have the highest sensitivity to rainfall-induced soil erosion and thus potential concomitant radiocesium transfers to downstream catchments and ultimately to the Pacific Ocean. The

high-est rainfall erosivity in these areas occurs during the trop-ical cyclone season (July–September). Monitoring radioce-sium transfers after major events will be important. In other catchments that may be impacted by future radiocesium con-tamination it will be important to similarly investigate the climatic context, as precipitation strongly influences post-fallout radiocesium dynamics.

The Supplement related to this article is available online at doi:10.5194/hess-20-2467-2016-supplement.

Acknowledgements. We would like to thank the reviewers for their constructive comments that helped improve the manuscript. This research was funded by the ANR (French National Research Agency) in the framework of the AMORAD project (ANR-11-RSNR-0002).

Edited by: N. Romano

References

Akaike, H.: A new look at the statistical model identification, Au-tomatic Control, IEEE Transactions on, 19, 716–723, 1974. Barry, R. G. and Chorley, R. J.: Atmosphere, weather and climate,

Routledge, 536 pp., 2009.

Belyaev, V. R., Golosov, V. N., Ivanova, N. N., Markelov, M. V., and Tishkina, E. V.: Human-accelerated soil redistribution within an intensively cultivated dry valley catchment in southern European Russia, IAHS-P, 291, 11–20, 2005.

Capolongo, D., Diodato, N., Mannaerts, C. M., Piccarreta, M., and Strobl, R. O.: Analyzing temporal changes in cli-mate erosivity using a simplified rainfall erosivity model in Basilicata (southern Italy), J. Hydrol., 356, 119–130, doi:10.1016/j.jhydrol.2008.04.002, 2008.

Chartin, C., Evrard, O., Onda, Y., Patin, J., Lefèvre, I., Ottlé, C., Ayrault, S., Lepage, H., and Bonté, P.: Tracking the early dispersion of contaminated sediment along rivers draining the Fukushima radioactive pollution plume, Anthropocene, 1, 23– 34, 2013.

Chartin, C., Evrard, O., Onda, Y., Laceby, P., Cerdan, O., Lep-age, H., Ottlé, C., Lefèvre, I., and Bonté, P.: The impact of ty-phoons on sediment connectivity: Lessons learnt from contami-nated coastal catchments of Fukushima Prefecture (Japan), Earth Surf. Process. Landf., in revision, 2016.

Chino, M., Nakayama, H., Nagai, H., Terada, H., Katata, G., and Yamazawa, H.: Preliminary estimation of release amounts of131I and137Cs accidentally discharged from the Fukushima Daiichi nuclear power plant into the atmosphere, J. Nucl. Sci. Technol., 48, 1129–1134, 2011.

Daly, C., Gibson, W. P., Taylor, G. H., Johnson, G. L., and Pasteris, P.: A knowledge-based approach to the statistical mapping of cli-mate, Clim. Res., 22, 99–113, 2002.

de Brogniez, D., Ballabio, C., Stevens, A., Jones, R., Montanarella, L., and van Wesemael, B.: A map of the topsoil organic carbon content of Europe generated by a generalized additive model, Eu-rop. J. Soil Sci., 66, 121–134, 2015.

Duan, W., He, B., Takara, K., Luo, P., Hu, M., Alias, N. E., and Nover, D.: Changes of precipitation amounts and extremes over Japan between 1901 and 2012 and their connection to climate indices, Clim. Dynam., 45, 2273–2292, 2015.

Eltaif, N. I., Gharaibeh, M. A., Al-Zaitawi, F., and Alhamad, M. N.: Approximation of Rainfall Erosivity Factors in North Jordan, Pedosphere, 20, 711–717, doi:10.1016/S1002-0160(10)60061-6, 2010.

ESRI: ArcGIS Desktop: Release 10, Redlands, Environmental Sys-tems Research Institute, CA, 2011.

Evrard, O., Chartin, C., Onda, Y., Patin, J., Lepage, H., Lefèvre, I., Ayrault, S., Ottlé, C., and Bonté, P.: Evolution of radioactive dose rates in fresh sediment deposits along coastal rivers drain-ing Fukushima contamination plume, Scientific reports, 3, 3079, doi:10.1038/srep03079, 2013.

Evrard, O., Chartin, C., Onda, Y., Lepage, H., Cerdan, O., Lefevre, I., and Ayrault, S.: Renewed soil erosion and remobilisation of radioactive sediment in Fukushima coastal rivers after the 2013 typhoons, Sci. Rep., 4, 4574, doi:10.1038/srep04574, 2014. Evrard, O., Laceby, J. P., Lepage, H., Onda, Y., Cerdan, O.,

and Ayrault, S.: Radiocesium transfer from hillslopes to the Pacific Ocean after the Fukushima Nuclear Power Plant accident: A review, J. Environ. Radioactiv., 148, 92–110, doi:10.1016/j.jenvrad.2015.06.018, 2015.

García-Oliva, F., Maass, J. M., and Galicia, L.: Rainstorm Anal-ysis and Rainfall Erosivity of a Seasonal Tropical Region with a Strong Cyclonic Influence on the Pacific Coast of Mexico, J. Appl. Meteorol., 34, 2491–2498, doi:10.1175/1520-0450(1995)034<2491:RAAREO>2.0.CO;2, 1995.

Hastie, T. and Tibshirani, R.: Generalized additive models, Statisti-cal science, 297–310, 1986.

Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G., and Jarvis, A.: Very high resolution interpolated climate surfaces for global land areas, Int. J. Climatol., 25, 1965–1978, 2005.

Jarvis, A., Reuter, H. I., Nelson, A., and Guevara, E.: Hole-filled SRTM for the globe Version 4, available from the CGIAR-CSI SRTM 90m Database http://srtm.csi.cgiar.org (last access: 15 April 2016), 2008.

JMA (Japanese Meteorological Agency) RSMC Tokyo – Ty-phoon Center: http://www.jma.go.jp/jma/jma-eng/jma-center/ rsmc-hp-pub-eg/RSMC_HP.htm, last access: 12 February 2016. Karki, K. B. and Shibano, H.: Soil loss in a forested watershed un-derlain by deeply weathered granite: comparison of observations to predictions of a GIS-based USLE, Bulletin of the Tokyo Uni-versity Forests, 2006.

Karydas, C. G., Tzoraki, O., and Panagos, P.: A New Spatiotempo-ral Risk Index for Heavy Metals: Application in Cyprus, Water, 7, 4323–4342, 2015.

Kato, H., Onda, Y., and Teramage, M.: Depth distribution of 137Cs, 134Cs, and 131I in soil profile after Fukushima Dai-ichi Nu-clear Power Plant Accident, J. Environ. Radioactiv., 111, 59–64, doi:10.1016/j.jenvrad.2011.10.003, 2012.

Kim, S., Nakakita, E., Tachikawa, Y., and Takara, K.: Precipitation changes in Japan under the A1B climate change scenario, Ann. J. Hydr. Eng., 54, 127–132, 2010.

Kinnell, P. I. A.: Event soil loss, runoff and the Universal Soil Loss Equation family of models: A review, J. Hydrol., 385, 384–397, doi:10.1016/j.jhydrol.2010.01.024, 2010.

Kitamura, A., Yamaguchi, M., Kurikami, H., Yui, M., and On-ishi, Y.: Predicting sediment and cesium-137 discharge from catchments in eastern Fukushima, Anthropocene, 5, 22–31, doi:10.1016/j.ancene.2014.07.001, 2014.

Larionov, G.: Erosion and deflation of soils, Moscow University Press, Moscow, Russia, 200 pp., 1993.

Lee, J.-H. and Heo, J.-H.: Evaluation of estimation methods for rainfall erosivity based on annual precipitation in Korea, J. Hy-drol., 409, 30–48, doi:10.1016/j.jhydrol.2011.07.031, 2011. Lu, H. and Yu, B.: Spatial and seasonal distribution of rainfall

ero-sivity in Australia, Soil Research, 40, 887-901, 2002.

Ma, X., He, Y., Xu, J., van Noordwijk, M., and Lu, X.: Spatial and temporal variation in rainfall erosivity in a Himalayan watershed, CATENA, 121, 248–259, doi:10.1016/j.catena.2014.05.017, 2014.

McFarlane, D., Westcott, T., and Davies, J.: Rainfall erosivity in Western Australia, Hydrology and Water Resources Symposium 1986: River Basin Management, Preprints of Papers, 350, 1986. Mello, C. R., Viola, M. R., Beskow, S., and Norton, L. D.: Multi-variate models for annual rainfall erosivity in Brazil, Geoderma, 202–203, 88–102, doi:10.1016/j.geoderma.2013.03.009, 2013. Meusburger, K., Steel, A., Panagos, P., Montanarella, L., and

Alewell, C.: Spatial and temporal variability of rainfall erosiv-ity factor for Switzerland, Hydrol. Earth Syst. Sci., 16, 167-177, doi:10.5194/hess-16-167-2012, 2012.

Moore, T.: Rainfall erosivity in East Africa, Phys. Geogr., 147–156, 1979.

Mouri, G., Golosov, V., Shiiba, M., and Hori, T.: Assessment of the caesium-137 flux adsorbed to suspended sediment in a reservoir in the contaminated Fukushima region in Japan, Environ. Pollut., 187, 31–41, doi:10.1016/j.envpol.2013.12.018, 2014.

Nagao, S., Kanamori, M., Ochiai, S., Tomihara, S., Fukushi, K., and Yamamoto, M.: Export of134Cs and 137Cs in the Fukushima river systems at heavy rains by Typhoon Roke in September 2011, Biogeosciences, 10, 6215–6223, doi:10.5194/bg-10-6215-2013, 2013.

Nazarok, P., Kruglov, O., Menshov, O., Kutsenko, M., and Sukho-rada, A.: Mapping soil erosion using magnetic susceptibility. A case study in Ukraine, Solid Earth Discuss., 6, 831–848, doi:10.5194/sed-6-831-2014, 2014.

Oliveira, P. T. S., Wendland, E., and Nearing, M. A.: Rain-fall erosivity in Brazil: A review, CATENA, 100, 139–147, doi:10.1016/j.catena.2012.08.006, 2013.

Onishi, Y., Kurikami, H., and Yokuda, S.: Simulation of Sediment and Cesium Transport in the Ukedo River and the Ogi Dam Reservoir during a Rainfall Event using the TODAM Code, Pa-cific Northwest National Laboratory (PNNL), Richland, WA, 124 pp., 2014.

Peel, M. C., Finlayson, B. L., and McMahon, T. A.: Updated world map of the Köppen-Geiger climate classification, Hy-drol. Earth Syst. Sci., 11, 1633–1644, doi:10.5194/hess-11-1633-2007, 2007.

Poggio, L., Gimona, A., and Brewer, M. J.: Regional scale mapping of soil properties and their uncertainty with a large number of satellite-derived covariates, Geoderma, 209, 1–14, 2013. Renard, K. G. and Freimund, J. R.: Using monthly precipitation

data to estimate the R factor in the revised USLE, J. Hydrol., 157, 287–306, doi:10.1016/0022-1694(94)90110-4, 1994. Renard, K. G., Foster, G. R., Weesies, G. A., and Porter, J. P.:

RUSLE: Revised universal soil loss equation, J. Soil Water Con-serv., 46, 30–33, 1991.

Renard, K. G., Foster, G. R., Weesies, G. A., McCool, D., and Yo-der, D.: Predicting soil erosion by water: a guide to conservation planning with the revised universal soil loss equation (RUSLE), Agriculture Handbook Washington, 403 pp., 1997.

Rosa, D. D. L., Crompvoets, J., Mayol, F., and Moreno, J. A.: Land vulnerability evaluation and climate change impacts in Andalu-cia, Agrophys. J., 10, 225–238, 1996.

Saito, K. and Onda, Y.: Outline of the national mapping projects im-plemented after the Fukushima accident, J. Environ. Radioactiv., 139, 240–249, 2015.

Saito, K., Tanihata, I., Fujiwara, M., Saito, T., Shimoura, S., Otsuka, T., Onda, Y., Hoshi, M., Ikeuchi, Y., and Takahashi, F.: Detailed deposition density maps constructed by large-scale soil sampling for gamma-ray emitting radioactive nuclides from the Fukushima Dai-ichi Nuclear Power Plant accident, J. Environ. Radioactiv., 139, 308–319, 2015.

Sawhiney, B.: Selective sorption and fixation of cations by clay min-erals: a review, Clays Clay Miner, 20, 93–100, 1972.

Shamshad, A., Azhari, M. N., Isa, M. H., Hussin, W. M. A. W., and Parida, B. P.: Development of an appropriate procedure for estimation of RUSLE EI30 index and preparation of erosivity maps for Pulau Penang in Peninsular Malaysia, CATENA, 72, 423–432, doi:10.1016/j.catena.2007.08.002, 2008.

Shiono, T., Ogawa, S., Miyamoto, T., and Kameyama, K.: Expected impacts of climate change on rainfall erosivity of farmlands in Japan, Ecol. Eng., 61, 678–689, 2013.

Svetlitchnyi, A.: Soil Erosion Induced Degradation of Agroland-scapes in Ukraine: Modeling, Computation and Prediction in Conditions of the Climate Changes, in: Regional Aspects of Climate-Terrestrial-Hydrologic Interactions in Non-boreal East-ern Europe, edited by: Groisman, P., and Ivanov, S., NATO Sci-ence for Peace and Security Series C: Environmental Security, Springer Netherlands, 191–199, 2009.

Takeuchi, B. K. and Ishidaira, H.: The examination of sediment pro-duction estimation by the USLE method in mountainous basins of Japan, Hydr. Eng., 45, 85–90, 2001.

Tamura, T.: Consequences of activity release: selective sorption re-actions of cesium with soil minerals, Nucl. Safety 5, 262–268, 1964.

Thakur, P., Ballard, S., and Nelson, R.: An overview of Fukushima radionuclides measured in the north-ern hemisphere, Sci. Total Environ., 458–460, 577–613, doi:10.1016/j.scitotenv.2013.03.105, 2013.

Tran, T. A., Mitani, Y., Ikemi, H., and Matsuki, H.: Human Impacts on Erosion and Deposition in Onga River Basin, Kyushu, Japan, Memoirs of the Faculty of Engineering, Kyushu University, 71, 47-65, 2011.

USDA: Rainfall Intensity Summarization Tool (RIST), United States Department of Agriculture (USDA), Agriculture Research Service, National Sedimentation Laboratory, Oxford, Missis-sippi, Version 3.88, 2013.

Wang, X. L., Chen, H., Wu, Y., Feng, Y., and Pu, Q.: New Tech-niques for the Detection and Adjustment of Shifts in Daily Pre-cipitation Data Series, J. Appl. Meteorol. Climatol., 49, 2416– 2436, doi:10.1175/2010JAMC2376.1, 2010.

Wikipedia Pacific typhoon climatology: https://en.wikipedia.org/ wiki/Pacific_typhoon_climatology, last access: 13 February 2016.

Wilkinson, S. N., Prosser, I. P., Rustomji, P., and Read, A. M.: Mod-elling and testing spatially distributed sediment budgets to relate erosion processes to sediment yields, Environ. Model. Softw., 24, 489–501, 2009.

Wischmeier, W. H. and Smith, D. D.: Rainfall energy and its re-lationship to soil loss, T. Am. Geophys. Union, 39, 285–291, doi:10.1029/TR039i002p00285, 1958.

Wischmeier, W. H. and Smith, D. D.: Predicting rainfall erosion losses-A guide to conservation planning, Predicting rainfall ero-sion losses – A guide to conservation planning, 58 pp., 1978. Wood, S. N.: mgcv: GAMs and generalized ridge regression for r,

R news, 1, 2–25, 2001.

Yamaguchi, M., Kitamura, A., Oda, Y., and Onishi, Y.: Predict-ing the long-term 137Cs distribution in Fukushima after the Fukushima Dai-ichi nuclear power plant accident: a parame-ter sensitivity analysis, J. Environ. Radioactiv., 135, 135–146, doi:10.1016/j.jenvrad.2014.04.011, 2014.

Yamashiki, Y., Onda, Y., Smith, H. G., Blake, W. H., Wakahara, T., Igarashi, Y., Matsuura, Y., and Yoshimura, K.: Initial flux of sediment-associated radiocesium to the ocean from the largest river impacted by Fukushima Daiichi Nuclear Power Plant, Sci. Reports, 4, doi:10.1038/srep03714, 2014.

Yoshimura, K., Onda, Y., and Kato, H.: Evaluation of radio-caesium wash-off by soil erosion from various land uses using USLE plots, J. Environ. Radioactiv., 139, 362–369, doi:10.1016/j.jenvrad.2014.07.019, 2015.

Yu, B., Hashim, G., and Eusof, Z.: Estimating theRfactor with lim-ited rainfall data: a case study from peninsular Malaysia, Journal of Soil and Water Conservation, 56, 101–105, 2001.